Test_results 
The largest data set tested with the software to 
date is a topographic contour overlay of a USGS 
1:100,000-scale quadrangle. This data set 
contained more than 460 million pixels, of which 
6.6 million were linework. (The output vector 
data set, stored in a relatively compressed binary 
format, was about 1.7 Mb. The vectorization 
process took about 85 minutes on a 17-mip 
workstation.) 
In this test 111 of the 256 possible neighborhood 
states were represented. However, only 11 
neighborhood states accounted for 96 percent of 
the actual neighborhoods, and 19 states accounted 
for 99 percent of the actual neighborhoods. 
  
/* prototypes of functions that comprise the 
rules base */ 
void rl(),r2(), /* ....*/ rl6(); 
/* array of 256 pointers to functions, 
statically declared and initialized */ 
void (*ruleFunc[256])() = { 
rl, r3, r2, r2, r2, r7, r2, r2, r2, r7, 
/* 240 more assignments go here */ 
rló, rló, rl6, rl6, rl16, ri16 
32 
/* program fragment to illustrate call to a 
rule function */ 
main() 
{ 
int currState; 
/* many lines of code here */ 
/* evaluation of neighborhood state */ 
currState - evaluate(); 
/* fire appropriate rule */ 
(*ruleFunc[currState])(); 
/* many more lines of code here */ 
J 
  
  
  
Figure 4 Implementation of the rule set as an 
array of pointers to functions. 
Analysis of this and other test data sets suggests 
the 16 rules identified in this paper define a 
reasonable cartographic vectorization process. 
However, the analysis also suggests that these 
rules are not uniquely correct. Characteristics 
of the output data can be changed relatively 
easily by altering the rules. The speed and 
efficiency can likely be improved by extending the 
rule set to deal better with certain special 
cases. 
CONCLUSIONS 
This research developed work of previous 
investigators into a prototype software system. 
Cartographic vectorization requires only one 
sequential pass through the raster data set. 
Acceptable processing speeds do not require that 
large amounts of the raster data be held in memory 
at any one time. 
The nature of the vector output can be controlled 
by a set of 16 rules. These rules can be coded in 
software in a manner that makes them easy to 
modify. 
The 16 rules shown in table 1 work well on 
cartographic data from USGS quadrangles. However, 
these rules are probably not uniquely correct. 
The behavior of the system can be modified by 
changing these rules. It is probable that more 
intelligence could be built into an extended rule 
set to alter system behavior and improve 
performance. 
The combination of sequential processing and 
rule-based decision making was quite effective in 
this application. That combination may be 
34 
applicable to other aspects of cartographic raster 
processing such as raster line skeletonization. 
REFERENCES 
Agrawala, Ashok K. and Ashok V. Kulkarni, 1977. A 
sequential approach to the extraction of shape 
features. Computer Graphics and Image Processing, 
6, pp. 538-577. 
Douglas, David H. and Peucker, Thomas K., 1973. 
Algorithms for the reduction of the number of 
points required to represent a digitized line or 
its caricature. The Canadian Cartographer, 10(2), 
pp. 112-122. 
Golay, M.J.E., 1969. Hexagonal parallel pattern 
transformations. IEEE Transactions on Computers, 
C-18(8). 
Greenlee, David D., 1987. Raster and vector 
processing for scanned linework.  Photogrammetric 
Engineering and Remote Sensing, 55(10), pp. 
1383-1387. 
Peuquet, Donna J., 1981. An examination of 
techniques for reformatting digital cartographic 
data/part 1: the raster-to-vector process. 
Cartographica, 18(1), pp. 34-48. 
Rosenfeld, Azriel and Pfaltz, John L., 1966. 
Sequential operations in digital picture 
processing. Journal of the Association for 
Computing Machinery, 13(4), pp. 471-494. 
c 
ct o 0 «dob "dd 53 O U 
C. 0o O c o Oo